x-dm

name	x-dm
description	Explicitly invoked Oparax workflow for sending already-prepared X DMs sequentially until the queue ends or X refuses a send. Never invoke implicitly.

X-dm

Use only when Farzan explicitly invokes $x-dm. That invocation authorizes sending the exact prepared messages for queued x_av records.

Perform the complete workflow inline in the current task. Never delegate it or start a subagent. Use only the Codex in-app Browser and .codex/outreach/outreach.py. Keep all X work sequential.

At the start, establish one persistent Codex goal: exhaust the x_av queue or stop on a defined X hard-stop. Reuse the active goal on automatic continuation; never create parallel goals or workers. A per-turn runtime boundary is not a stopping condition: leave the goal active and continue from the next x_av record automatically. Complete the goal only when next send returns null or a defined hard-stop is reached.

Loop:

1. Run rtk python3 .codex/outreach/outreach.py next send.
2. If the result is null, stop normally; the queue is exhausted.
3. In X Messages, start a new message, search the exact returned handle, and select only an exact handle match.
4. If the exact returned message is already visibly present, run rtk python3 .codex/outreach/outreach.py resolve <handle> x_done and continue without resending.
5. If the account is greyed out or has no usable composer, run rtk python3 .codex/outreach/outreach.py resolve <handle> x_unav and continue.
6. Otherwise, send the exact returned message without changing any character.
7. Count success only after the complete message remains visible and the composer is empty. Then run rtk python3 .codex/outreach/outreach.py resolve <handle> x_done and continue.

Hard-stop immediately if X shows Failed to send message, Failed, Try Again, a warning, rate limit, locked session, changed UI, recipient mismatch, or an unverifiable send. Do not retry, inspect another handle, or update the current record. Tell Farzan that X is refusing sends and identify the unchanged handle.

Do not produce a final response while next send returns a record. Do not stop after an arbitrary batch or because one turn reaches its runtime boundary. Continue in the same task until the queue is empty or a hard-stop condition occurs.

Before reporting, run rtk python3 .codex/outreach/outreach.py count send. Never open LeanSpark. Return newly sent handles, unavailable outcomes, the stopping reason, and that exact remaining x_av count.